Google AI gets cagey when asked who won the 2020 elections
Separate tests found evasive answers about the U.S. presidential result; Google said it was fixing a verification gap.
TL;DR
- Ars Technica and The Atlantic found that Google AI Mode sometimes avoided naming the 2020 election winner.
- Google said a technical gap affected verification of the 2020 result and 2022 Senate races.
- The Atlantic later received a direct answer in a subsequent test.
Ars Technica reported that Google AI Mode referred a question about the 2020 U.S. presidential election to official sources instead of directly naming the winner. The Atlantic separately obtained a similarly indirect answer in its own test. [1] [2]
The certified winner was Joe Biden. Ars Technica found that the elections receiving indirect answers differed among its staff, so the observed behavior was not uniform across users. [1]
A Google spokesperson told both publications that a technical gap in the system for verifying election responses affected the 2020 election and 2022 Senate races. Google said it was working to fix the issue. [1] [2]
The Atlantic reported that a later test returned Biden’s name directly. That single later result does not establish that the issue was resolved for every user or every election query. [2]
Why it matters
The incident connects AI product reliability with election information, where a missing answer can be consequential even when official results are available. The reports also show why a fix needs to be assessed across queries and users.
Editor's note
Both outlets ran their own tests; responses varied. Google attributed the problem to a technical gap, and neither report establishes deliberate censorship or a universally completed fix.